murra181/Tanzania-water-well-challenge

Project 3 flatiron # Tanzania Water Challenge - Tanzania has a severe water crisis that is affecting over half the country. 4 million people do not have access to safe water and 30 million of the countries 57 million do not have improved sanitation. This is the classification challange from drivendata.org. We will be looking to see which water wells need our attention first and be able to predict which ones going forward. ## Business Case - Using data provided from drivendata.org for the "Pump it Up: Data Mining the Water Table" Challenge I will be building a model to predict if a well is functional, functional but in the need of a repair, or not functional. Accuracy is what is being judged so this is what my models will be trained for. ## EDA - Looked through each column to see what had to many unique variables what was viable and what were mere copies or generalized forms of other columns. I reduced the columns from thirty nine independent variables to just twenty and got rid of 4,788 rows that were missing data. Categorical features were then dummied and readied for modeling ## Baseline models - We will be looking at accuracy for four base models as they are what makes the most sense for a ternary classification and the data we have. Decision Tree: 74.56% Random Forest: 78.55% Adaboost: 72.14% Bagged Trees: 78.6% Random forest and bagged trees are very similar so it makes sense that they are so close. I took the Random Forest model and Hyperparaeter Tuned it. ## Final Model - GridSearchCV was used on the Random Forest Model an increased the accuracy by 1.16% to 79.61% I then Bagged the model and increased it to 80.11% ## Conclusion - The model was accurately able to predict whether a pump was was functional, functional but needs repaired or not functional at 80.11% This is what the competition was looking for but the biggest part that I think needs improving for this data to be useful is that 13.75% of wells predicted as functional were not. ## Future Work - Remove o …